Share story
AI

Google revives diffusion research as open-weight DiffusionGemma model

Google released DiffusionGemma, an open-weight Gemma model under the Apache 2 license that returns the company's Gemini Diffusion research to public availability, developer Simon Willison reported. Google briefly released the experimental Gemini Diffusion model last May and made no further announcements about it until now. NVIDIA is hosting the model, google/diffusiongemma-26B-A4B-it, for free on its NIM cloud API. Willison's test of the hosted model returned 2,409 tokens in 4.4 seconds, at least 500 tokens per second. His earlier preview test of the experimental Gemini Diffusion model ran at 857 tokens per second.
Sources
Simon Willison's Weblog
In this story
Google Nvidia
Published by Tech & Business, a media brand covering technology and business. This story was sourced from Simon Willison's Weblog and reviewed by the T&B editorial agent team.
Back to Newswire
Keep reading
Full wire
AI Infrastructure
AI Infrastructure

Google previews managed Gemini search for air-gapped data centers

Google announced a preview of managed enterprise search with Gemini on Google Distributed Cloud air-gapped, allowing regulated organizations to query files without connecting to the public internet or Google's global infrastructur...

Cloud Blog
AI Infrastructure
AI Infrastructure

Google open-sources TorchTPU for native PyTorch workloads on TPUs

Google released TorchTPU as open source on GitHub, saying the stack lets developers train and serve existing PyTorch models on its TPU accelerators with minimal code changes. Co-developed with Meta's PyTorch team, it supports fami...

Cloud Blog
AI
AI

Google opens preview of Gemini agents for government workflows

Google Cloud has made Gemini for government services available in preview, offering agencies agents for tasks including public-records discovery, document redaction, bid evaluation and grant expense audits. Google says connectors...

Cloud Blog
AI Infrastructure
AI Infrastructure

llm-d 1.0 adds traffic priorities and upgrade controls for model serving

Google announced the release of llm-d 1.0, an open-source stack that coordinates model servers across Kubernetes clusters. The release adds stable v1 interfaces, router high availability and controls that let interactive requests ...

Cloud Blog